Stable calcium-free myocilin olfactomedin domain variants reveal challenges in differentiating between benign and glaucoma-causing mutations

Nonsynonymous gene mutations can be beneficial, neutral, or detrimental to the stability, structure, and biological function of the encoded protein, but the effects of these mutations are often not readily predictable. For example, the β-propeller olfactomedin domain of myocilin (mOLF) exhibits a complex interrelationship among structure(s), stability, and aggregation. Numerous mutations within mOLF are linked to glaucoma; the resulting variants are less stable, aggregation-prone, and sequestered intracellularly, causing cytotoxicity. Here, we report the first stable mOLF variants carrying substitutions in the calcium-binding site that exhibit solution characteristics indistinguishable from those of glaucoma variants. Crystal structures of these stable variants at 1.8–2.0-Å resolution revealed features that we could not predict by molecular dynamics simulations, including loss of loop structure, helix unwinding, and a blade shift. Double mutants that combined a stabilizing substitution and a selected glaucoma-causing single-point mutant rescued in vitro folding and stability defects. In the context of full-length myocilin, secretion of stable single variants was indistinguishable from that of the WT protein, and the double mutants were secreted to varying extents. In summary, our finding that mOLF can tolerate particular substitutions that render the protein stable despite a conformational switch emphasizes the complexities in differentiating between benign and glaucoma-causing variants and provides new insight into the possible biological function of myocilin.

The effect of a nonsynonymous coding mutation on a protein can be beneficial, benign, or detrimental to downstream stability, structure, cellular trafficking, and biological function but often not in a readily predictable manner. Disease-associated proteins are commonly associated with a loss of function due to, for example, a reduction in biological activity or enhanced cellular degradation, but gain-of-toxic function, such as aggregation and cell death, is also well-established (1). At the protein level, evolutionary adaptation is for function, not overall protein stability (2,3), as systematic studies of enzymes have revealed that charged or polar residues in sequestered active sites are optimized for catalysis (4) with compensatory mutations introduced elsewhere to balance stability (5). In cells, the molecular chaperone network detects client defects during folding and triages such proteins to enable refolding or degradation (6). Finally, the protein engineering field attempts to exploit the interplay between biophysical stability and function to improve upon or generate new biotechnologically relevant biocatalysts (7).
The ϳ30-kDa five-bladed ␤-propeller (8) olfactomedin domain of myocilin (mOLF) 2 (Fig. 1, A and B) is sensitive to mutations, many of which are causative for heritable forms of glaucoma (9). Disease-associated mOLF variants are destabilized compared with wildtype (WT) (10,11), facilitate amyloid aggregation in vitro under physiological conditions (12,13), and confer an aberrant interaction with the molecular chaperone Grp94 in cells (14). Ultimately, this leads to cytotoxicity and cell death (15) in the trabecular meshwork (TM) extracellular matrix, a key diseased ocular tissue associated broadly with glaucoma (16,17). WT mOLF exhibits thermal stability typical of human proteins near 52°C (18), but it too can be driven to aggregate by accessing a partially folded state via slightly elevated temperature, agitation, low pH, detergent, acid, or oxidizing or reducing agent (12,13).
In contrast to our molecular comprehension of the role of mutations in myocilin in eliciting a glaucoma phenotype, no discrete biological function has yet been assigned to WT myocilin since its discovery 20 years ago. The diverse olfactomedin protein family is found across chordates with numerous (19,20) subfamilies harboring different modular domains. For example, gliomedin possesses a collagen domain: interaction of its trimerized OLF domain with neuronal cell adhesion molecule (21) and likely other proteins (22) aids in the molecular assembly of the nodes of Ranvier in the peripheral nervous system.
We previously observed that the Mus musculus olfactomedin domain of collomin (cOLF; ϳ25% sequence identity to mOLF; Fig. S1; 1.0-Å root mean square deviation (r.m.s.d.); Ref. 23), which falls into the gliomedin subfamily of olfactomedins and possesses a gene organization similar to the olfactomedins found in invertebrates like nematodes (19), is 20°C more thermally stable (Table 1) and significantly more resistant to denaturation than mOLF (23). Why has nature elected to employ an aggregation-prone, disease-associated mOLF in the TM of the human eye where mutations, sustained UV exposure, and other environmental and mechanical stressors (17, 24 -26) likely render myocilin susceptible to aggregation when a more stable variant like cOLF could, in theory, be resistant to these stressors and avoid disease (23)? Upon inspection of the major distinguishing structural feature in mOLF compared with cOLF, namely, the presence of a central heptacoordinate calcium ion bound in the hydrophilic central pore (Fig. 1, A and B), we considered the counterintuitive suggestion that the missing ionic interaction endows cOLF with higher stability.
Here, we probed the properties conferred by the metal-binding site on mOLF stability, structure, and cellular trafficking. Although previous work from our group indicated that the calcium site was a liability to stability (27), here we show that other mutations in the central ion-binding site are either neutral or increase thermal stability. These stable variants adopt an alternative conformation that we could not simulate in silico, with significant loop disorder and structural rearrangements in the first complete blade, including unwinding of the side helix. The variants are still secreted from cells when present in the context of full-length myocilin and can offset, at least to some extent, folding and secretion defects associated with selected glaucoma-causing mOLF variants. Collectively, our study offers insight into the folding and evolution of the OLF ␤-propeller, the relative importance of stability and structure for interactions with molecular chaperones, and the interrelationship among protein stability, native structure, and function.

Calcium-ablating mOLF variants at position 380, but not at 478, are thermally compromised
In line with the nearly 40 individual mutations we introduced previously into mOLF, which had either a neutral or detrimental effect on thermal stability (8,10,11,27), our knowledge of the effect of the calcium-binding site on mOLF stability and structure (Fig. 1, A and B) to date has been limited to the effect of the glaucomatous mOLF(D380A) variant, which abolished Ca 2ϩ binding at the expense of ϳ7°C stability (melting temperature (T m ) ϭ 46.6°C) (27). We had not assessed whether calcium removal depends on the particular substitution, e.g. D380A versus D380N (28), or the specific calcium ligand. However, we noticed that calcium ligands Asp-380 and Asn-428 in mOLF are largely conserved across the olfactomedin protein family, whereas Asp-478 exhibits larger variation (Fig. S2) (8). In thermostable cOLF, residues at the equivalent position of Asp-478 have side chains that cannot coordinate metals, for example alanine in human cOLF and serine in mouse cOLF ( Fig. S2) (23).
Point mutants to introduce Ala, Ser, or Asn in place of Asp-380 or Asp-478 (Table 1) were generated by following established procedures (see "Experimental procedures"). In line with the results for D380A (27), changing Asp-380 to either Ser or Asn abolished calcium affinity, decreased thermal stability, and reduced soluble expression levels. Surprisingly, substitutions at Asp-478 abolished calcium binding as expected but conferred ϳ5-7°C higher thermal stability than WT mOLF. Our only prior successful attempt at stabilizing mOLF was using a computational design resulting in a T m of ϳ70°C, which involved 21 simultaneous mutations (mOLF (21-variant)) with an experimentally confirmed, intact calcium site (Ref. 29 and see below).

mOLF(D478S) rescues folding and stability of glaucomaassociated variants D380A, P370L, and Y437H
The combination of the stabilizing D478S variant with selected glaucoma variants D380A, Y437H, and P370L rescues their impaired stability (Table 1). mOLF(D380A/D478S) and mOLF(D380A/D478N) are ϳ10°C more stable than A, cartoon representation and structural features of WT mOLF structure (mOLF SNP E396D, PDB code 4WXS; dark pink) in top-down orientation. Blades labeled A-E are centered around a solvent-filled central pore with calcium-and sodium-binding sites. B, zoomed-in view of the calcium and sodium environment from A with coordinating calcium residues highlighted. Water molecules, exclusively sodium-coordinating residues, and main chain not involved in calcium coordination were omitted for clarity. Interactions Յ2.5 Å are shown as dashed lines. C, tertiary structure signature measured by near-UV CD overlaid with disease-associated variant D380A (data from Ref. 12) shows that Asp-380 and Asp-478 variants have nonnative tertiary structure similar to D380A. D, wavelength position of the peak emission ( max ) from excitation of intrinsic protein fluorescence reveals that Asp-380 and Asp-478 variants have a red-shifted maximum, suggesting exposed hydrophobic residues and nonnative structure. Error bars represent S.D. for two independent experiments. See also Figs. S1 and S2. deg, degrees. mOLF(D380A), promoting mOLF(D380A/D478S or D380A/ D478N) above WT mOLF stability where its aggregation propensity under physiological conditions is fully inhibited (not shown). To extend this observation, we generated additional double mutants, one harboring the severe phenotype variant P370L, which we have never been able to prepare in a folded state in Escherichia coli (11), and the other Y437H, a well-studied (30 -32) and highly aggregation-prone (13) variant. Like mOLF(D380A/ D478S), both mOLF(P370L/D478S) and mOLF(Y437H/D478S) were expressed and purified in a soluble form with reasonable yield. Thermal stabilities of these latter variants are in line with moderate glaucomatous single variants like D380A and T377M. In sum, whereas loss of calcium binding due to mutation of Asp-380 is destabilizing to mOLF, calcium ablation is not always destabilizing and in fact can be used to rescue biophysical defects of even the most deleterious glaucoma-associated variants.

Ensemble refinement of WT mOLF structures and molecular dynamics (MD) simulations of calcium-depleted mOLF indicate modest structural fluctuations
Given the poorly predicted increase in thermal stability conferred by mutation of Asp-478 with publicly available mutant prediction servers (Table S1) and nonnative structure indicated by CD and intrinsic fluorescence, we employed MD simulations to gain insight into the structural changes that might accompany the ablation of the calcium-binding site with the introduction of Ser-478. Simulations of WT mOLF for 500 ns at T ϭ 350 K reveal that the overall architecture remains largely unchanged, as expected for a stable protein ( Figs. 2A and S3). The side helix (residues Ile-304 -Gln-309) unravels, but no structural changes are seen in the ␤-sheets. These results are in line with ensemble refinement (33) of our three WT mOLF crystal structures (8), which reveals fluctuations in three loops (residues Thr-261-Lys-266, Thr-290 -Asp-294, and Tyr-442-Ala-445; Fig. 2B and Table S2). Next, we compared these results with the scenario in which the calcium ion was removed from the WT mOLF structure. Unexpectedly, MD simulations indicate a rigidification of numerous loops. In the environs of the calcium site, we observed conformational changes to Asn-428, Asp-380, and Ala-429, presumably as an attempt to accommodate repulsion from the negatively charged, calcium-coordinating side chains remaining upon removing the ion; the motion of Asn-428 is most dramatic (Figs. 2, C and E, and S3B). Finally, we conducted MD simulations on the D478S variant by mutating the WT mOLF structure after removal of calcium (Figs. 2, D and E, and S3B). Excluding the increased motion observed for the loop composed of residues Thr-290 -Asp-294, as seen in the ensemble refinement of WT mOLF, the simulations again reveal a rigidification of loops, such as for the long loop (residues Ile-360 -Gly-375; Fig. 1A) on the top face of the propeller. Overall, ensemble refinement of WT mOLF crystal structures and MD simulations of WT mOLF and relevant variants indicate changes in flexibility of the loops but do not suggest gross structural rearrangements, at least on the sub-s time scale.

Partially folded but stable mOLF domain variants
with WT mOLF (r.m.s.d. of ϳ1.3 Å) but adopt an alternative conformation. In the mOLF(D478N) and mOLF(D478S) central pore, a metal ion is present, but not in the same position as the WT mOLF Ca 2ϩ or Na ϩ or cOLF Na ϩ (Fig. 3, A-C). Instead, for mOLF(D478N) and mOLF(D478S), an ion assigned as Na ϩ coordinates the side chain of Asp-380 as well as the carbonyl oxygen of Ile-477 and six water molecules, whereas for mOLF(D380A/D478S), ions are not coordinated (Fig. 3D). For all Asp-478 variant structures, blades B-E are superimposable, but for blade A, there is a ϳ2-Å  Table S2.

Partially folded but stable mOLF domain variants
shift of all four strands away from the central pore.  Fig. 3, D and E) and unraveling of the side helix (labeled side helix region 3, Ile-304 -Gln-309; Figs. 1A and 3, D-F). We note that the extent to which we could fit models to electron density in these mobile regions (Figs. 3E and S4, A-C) was inversely correlated with rank order of thermostability when comparing mOLF(D478N) or mOLF(D478S) with mOLF(D380A/D478S) ( Table 1). A major shift of Trp-270 from being nestled in a hydrophobic pocket near Leu-257, Phe-307, and Phe-487 to an exposed conformation along with changes to Tyr-267 and Trp-489 in the Asp-478 variants (Fig.  3, G and H) likely contributes to the observed nonnative CD spectra (Fig. 1C). MD simulations recapitulate this loss of electron density; if side helix region 3 (Ile-304 -Gln-309) is removed from WT mOLF, increased mobility is observed for the loops of blades A and E (Fig. 3I) along with the shift in blade A as seen for Asp-478 variants (Fig. S3C). Still, the structures do not provide a clear reason for higher thermostability of the Asp-478 variants compared with WT. The overall contact order is similar (not shown), and although significant increased flexibility is detected upon ensemble refinement of mOLF(D478N), mOLF(D478S), and mOLF(D380A/D478S) compared with the otherwise well-ordered WT mOLF and cOLF structures (Figs. 2B and 3, J-M, and Table S2), chemical intuition would dictate that the combination of the loss of ordered loops and helical secondary structure would correspond to a less, not more, thermostable protein.
For further comparison, we next solved the 1.9-Å-resolution structure of our most stable mOLF variant to date (29), which harbors 21 computationally predicted mutations (Table 2 and Figs. 4A and S4D). The two monomers present in the asymmetric unit are more similar to WT than the Asp-478 variants with an overall r.m.s.d. of just 0.7-0.8 Å, but they each also exhibit unexpected alternative conformations (Fig. 4B). The monomer assigned as chain A is fully modeled, and blades overlay well with WT mOLF, but ensemble analysis reveals flexibility in the same three loops seen as disordered or missing among the Asp-478 variant structures (Fig. 4C). The second monomer, chain B, lacks density for residues 261-264, and ensemble refinement supports thermal motions in the loops of blades D, E, and A ( Fig. 4D and Table  S2). We surmise that substitution of L492G and possibly N493H weaken the hydrophobic interactions in blade E (Fig. 4, E and F), which propagate to neighboring blades D and A.

Alternative conformation adopted by stable Asp-478 variants bolsters evolutionary relationship of OLF to six-bladed propellers
Previously, a phylogenetic relationship between the OLF propeller and six-bladed propellers was suggested based on close structural similarity for blades A-D with a six-bladed Kelch domain (34). Evolutionary classification of protein domains based on structure (35) identifies a six-bladed propeller (DUF4221; PDB code 3S9J), a domain of unknown function from the bacterium Bacteroides vulgatus ATCC 8482, at a branchpoint with OLFs (Fig. S5A). Although the overall r.m.s.d.
for the two domains remains ϳ3 Å, the ϳ2-Å shift of blade A observed in mOLF Asp-478 variants results in a visually better superposition to the six-bladed propeller (PDB ID 3S9J) than with WT mOLF (Fig. S5, B and C). Perhaps by adopting a more ancestral-like fold (36), namely closer to a six-bladed propeller, the stable mOLF Asp-478 variants exhibit increased thermostability even while ablating an ionic interaction and weakening certain intermolecular interactions as these features were added later as the propeller acquired new function.

Cellular secretion profiles of stable single-point Asp-478 variants are indistinguishable from WT myocilin and when combined with glaucoma-causing variants rescue secretion to varying degrees
Myocilin is a component of the TM extracellular matrix and is thus secreted from TM cells. Disease-associated mutations

. Crystal structure of thermally stable mOLF(21-variant) reveals distinct conformations for each monomer in the asymmetric unit. A, cartoon representation of monomer A (beige; blades labeled A-E) with location of amino acids (yellow sticks) that differ from WT mOLF.
Orange sphere, Ca 2ϩ ; cyan sphere, Na ϩ . B, superposition of monomer A, monomer B (gray), and WT mOLF (mOLF SNP E396D, PDB code 4WXS; dark pink as in Fig. 3; partially transparent). Monomer A is nearly indistinguishable from WT mOLF, whereas monomer B exhibits three loop changes, circled in pale yellow and numbered 1 and 2 as in Fig. 3 and 4 as in Fig. 2 involving Tyr-442 previously shown to have alternative conformations for WT mOLF (8). C and D, ensemble refinement, displayed as main-chain lines for each state, of monomer A (C) reveals a similar degree of flexibility as compared with WT mOLF (Fig. 2B) with some degree of flexibility for loop/helix region 1 (residues Thr-261-Lys-266), loop region 2 (residues Thr-290 -Asp-294), side helix region 3 (residues Ile-304 -Gln-309), and loop region 4 (residues Tyr-442-Ala-445). Ensemble refinement of monomer B (D) shows a longer stretch of flexibility between the N terminus and loop/helix region 1 (residues Thr-261-Lys-266) as revealed by a compromised molecular clasp in the initial refinement and additional flexibility for loop Asn-469 -Tyr-473, which is shifted in the Asp-478 variant structures (Fig. 3, B and C). E and F, enlarged region 1 from B. E, monomer A and WT mOLF share interactions for the molecular clasp in blade E. Monomer A has an additionalstacking interaction between His-493 and Tyr-471 that is not present in WT mOLF. F, in monomer B, the molecular clasp is compromised, indicated by loss of H-bonding, an alternative conformation for the first ␤-strand of blade E, and loss of electron density for loop/helix region 1 (residues Thr-261-Lys-266) leading to blade A.

Partially folded but stable mOLF domain variants
are instead sequestered intracellularly (15,30,37,38), specifically within the ER (39), not the cytosol (40). Misfolded myocilin is recognized by the molecular chaperone Grp94 as not competent for secretion, but instead of undergoing efficient degradation, aggregated mutant myocilin accumulates, leading to ER stress, TM cell death (8,15,41), and a cascade that results in early onset ocular hypertension, the causal risk factor for glaucoma.
We compared the cellular secretion profiles of WT and characterized mOLF calcium variants in the context of fulllength myocilin using a HEK293 transiently transfected cell culture model, probing both extracellular and intracellular levels of myocilin (Fig. 5). The stable variants, D478S and D478N, which adopt a non-WT structure (Fig. 3), are secreted as for WT myocilin, with only low levels of soluble and insoluble intracellular myocilin detected. Thus, the nonnative structure exhibited by the calcium-free variants is not recognized by chaperone machinery, allowing these proteins to proceed through the secretory pathway.
Next we examined the secretion profile of myocilin double mutants that rescued folding and thermal stability of glaucomacausing variants in the context of purified mOLF (see Table 1). Myocilin single-point variants D380A, P370L, and Y437H were sequestered intracellularly with aggregates that are insoluble in detergent as reported previously (30). Of the three double mutants tested, only secretion of D380A/D478N (Fig. 5A), which has WT-like mOLF stability, was rescued to near WT levels. By contrast, P370L/D478S and Y437H/D478S, which have lower overall stability, are secreted to a lesser degree (Fig. 5B).

Discussion
The OLF ␤-propellers comprise a structurally distinct fivebladed family; data in this study along with previous observations support the hypothesis that OLF evolved from more common six-bladed propellers. Olfactomedins are unusual in that they are only found in multicellular organisms, and although they are associated with numerous human diseases, explicit biological functions have not been assigned for most subfamily members (42). For myocilin, the high sensitivity of mOLF to destabilization and misfolding due to mutation is well-documented (11) and is causative for ocular hypertension and familial glaucoma (43); however, its function has remained elusive since its discovery 20 years ago. In our original mOLF structures, we trapped slightly different conformations of well-ordered loops in different crystal forms, but neither these structures nor computational analyses presented in this study predicted that mutation of Asp-478 would lead to a more stable, secreted variant, much less one adopting an alternative conformation with a high degree of new localized flexibility.

Figure 5. Cellular secretion assays of full-length myocilin variants reveal that stabilizing mutations can, to varying extents, rescue secretion of glaucoma-associated variants. Dot blot analyses (left panels) of secreted myocilin (Myoc) from spent media from transfected HEK293T cells is compared with
Western blot analyses of detergent-soluble and -insoluble lysates (right panels). A, secretion and intracellular aggregate profiles of stable variants (D478N and D478S); a moderately stable, disease-associated variant (D380A); and a combination variant (D380A/D478N), which is rescued to a large extent. B, secretion and intracellular aggregate profiles of a stable variant (D478S); severely destabilized, glaucoma-associated variants (P370L and Y437H); and combination variants (P370L/D478S and Y437H/D478S), which are rescued to lesser extents than D380A/D478N in A. Labeled molecular mass markers are in kDa. Ponceau staining (dot blot; data not shown) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (Western blotting) were used as loading controls.

Partially folded but stable mOLF domain variants
Our current study demonstrates the complex relationship among mutations, structure, stability, and aggregation and elevates the need to more closely probe the relationship between location of mutation and glaucoma pathogenesis. The effects of a mutation on mOLF stability and folding are more complex than we appreciated until now. Indeed, our ability to use chemical intuition in predicting the consequences of other mutations in mOLF is relatively limited. Our early attempts to correlate stability effects of mutation and location in the mOLF structure did not define clusters very well (8), and using this analysis we predicted that Asp-478 variants would behave as D380A, a wellstudied destabilized, aggregation-prone glaucoma variant. The disease association of any given myocilin mutation for glaucoma is not definitive because without a confirmed family history, variations in myocilin are not a common cause of glaucoma (44). Successful clinical implementation of therapeutic strategies under development, namely eliminating mutant myocilin by CRISPR (45) or promoting myocilin degradation through manipulation of ER chaperone machinery (14,46,47), will rely heavily on accurate assessments of likelihood of pathogenicity of an identified myocilin mutation.
Our study also raises the intriguing possibility that the helix unwinding observed for Asp-478 variants is accessible in WT mOLF upon a functionally relevant stimulus. Regulated helix unwinding, also termed helix cracking, is known in the context of signal modulation in other proteins (48 -50), including propagation via coiled coils (51), which are found in myocilin. In addition, large kinetic barriers can prevent facile interconversion between two stable conformations of a protein that require a trigger, like for hemagglutinin (52). In this study, we discovered that the mOLF calcium-binding site is not always a liability in terms of thermal stability and that the presence of central calcium-mediated ionic interactions imparts the helix-wound WT mOLF structure but at the expense of thermal stability. This observation is in line with the general observation that proteins of mesophilic origin are thought to be marginally stable because there is no selection advantage for higher stability, only function (53), suitable levels of plasticity (54), and regulation (2).
Perhaps extracellular myocilin is responsive to calcium levels, such as fluxes associated with TM contractility (17,55). Namely, myocilin may be spring-loaded, detecting mechanical stimuli in the form of calcium flux and converting this signal to a structural readout (56). Such mechanotransduction would be tightly controlled as it might affect multiple aspects of TM function and could involve a variety of extracellular proteins (57). Although in the laboratory calcium is only fully removed from purified WT mOLF under denaturing conditions (27) and Asp-380 and Asp-478 would experience charge repulsion when in close proximity, calcium depletion could result in an alternative metal ion center within the pore, for example, by shuffling the ligands binding Na ϩ and Ca 2ϩ . Our MD simulations indicated a shift for calcium ligands when Ca 2ϩ is removed, and we can envision Asp-380 switching from binding Ca 2ϩ to a buried Lys-423. This would further require movement of Tyr-371, which forms a cationinteraction with Lys-423 in our WT mOLF structure and is ϳ2-Å shifted in the D478S structure. Propagation of these motions to N-terminal surface loops as seen in mOLF(D478S) and MD simulations might increase the surface accessibility of essential residues to interact with functionally relevant binding partners. In support of this proposal, the N-terminal ϳ50 amino acids of mOLF share less similarity with cOLF than the remaining ϳ200 residues, and connecting loops are consistently 1-3 residues longer than those of cOLF (Figs. 3A and S1).
The suggestion that a D478S-like conformation is functionally relevant may explain why these variants are competent for secretion even though in solution their in vitro CD profiles are indistinguishable from disease variants. The full complement of molecular chaperones inspecting myocilin during folding in TM cells is not established, but our study demonstrates that achieving native structure of mOLF with well-buried hydrophobic residues is not a strict requirement for myocilin secretion. Rather, the varying extents of secretion for the double mutants D478N/D380A, D478S/P370L, and D478S/Y437H suggest that a thermal stability threshold exists for recognition by a molecular chaperone, perhaps as a correlate of corresponding aggregation propensity. Future detailed investigations of myocilin trafficking, for both WT and D478S-like conformations, are warranted to establish whether kinetics (58,59) or binding partners also influence secretion.
In sum, our study reveals a conformation of mOLF that is distinct from the destabilized conformation leading to amyloidogenic aggregation or a molten globule competent for folding (60). The thermostable, calcium-free mOLF single-point variants are secreted from cells like WT myocilin even though in vitro solution characterization resembles those of destabilized glaucoma variants. Partially unfolded protein variants are typically less stable than the well-folded native state, but examples exist where alternative conformations are functional. The alternative conformation we captured in our study by ablating calcium binding in mOLF is likely accessible even without mutation, upon a calcium flux or mechanical stimulus or upon binding a ligand or partner protein. These new structural insights bring us closer to identifying the biological function of this cryptic ␤-propeller.

Molecular biology
The parent WT mOLF plasmid was generated as reported previously with a maltose-binding fusion protein and factor Xa cleavage site (10). Variants of mOLF were prepared by sitedirected mutagenesis according to the manufacturer's protocol (QuikChange Lightening kit). Forward and reverse primers (Table S3) specific for each mOLF variant were designed using Agilent Technologies QuikChange Primer Design and purchased from MWG Operon. Fidelity of plasmid sequences was confirmed by DNA sequencing (MWG Operon).

Protein expression
Plasmids were transformed into Rosetta-Gami 2 competent cells onto agar plates containing 60 g/ml ampicillin and 34 g/ml chloramphenicol selective LB broth. Starter cultures were prepared by adding one bacterial colony of mOLF variant to 250 ml of LB medium containing selective ampicillin/chloramphenicol antibiotics followed by incubation at 37°C with

Partially folded but stable mOLF domain variants
shaking overnight (225-250 rpm). For large-scale cultures, 25 ml of the starter culture was used to inoculate each 1-liter solution of Superior broth in 2-liter baffled flasks, and the solutions were allowed to grow at 37°C in a shaking incubator (225-250 rpm) until the cells reached an optical density (OD) of ϳ1.5-1.7 measured at 600 nm. At this point, the temperature was reduced to 18°C, flasks were equilibrated for 1.5 h at 18°C, and protein expression was induced with 500 M isopropyl ␤-Dthiogalactopyranoside. The cells continued to incubate overnight at 18°C for 16 -18 h. The next day, the cells were harvested by centrifugation (2,380 ϫ g) with 10-min spins, flash frozen with liquid nitrogen, and stored at Ϫ80°C.

Purification of mOLF variants
MBP-mOLF variant cells (3 g) were resuspended in 10 ml of amylose wash buffer (10 mM Na 2 HPO 4 , 10 mM KH 2 PO 4 , 200 mM NaCl, 1 mM EDTA) with 1ϫ cOmplete protease inhibitor mixture (Roche Applied Science) and lysed by passage through a French press twice. Cell debris was removed by centrifugation at 161,716 ϫ g at 4°C for 45 min. The soluble fraction was loaded onto an amylose-affinity column (New England Biolabs) equilibrated with amylose wash buffer, and MBP-mOLF fusion protein was eluted with amylose elution buffer (10 mM Na 2 HPO 4 , 10 mM KH 2 PO 4 , 200 mM NaCl, 1 mM EDTA, 10 mM maltose). Final fractionation using a Superdex 75 (GE Healthcare) size-exclusion column equilibrated with gel filtration buffer (10 mM Na 2 HPO 4 , 10 mM KH 2 PO 4 , 200 mM NaCl) isolated monomeric MBP-mOLF variants. MBP-mOLF variants were then cleaved with factor Xa (New England Biolabs) in a 50:1 protein-to-factor Xa reaction solution for 18 h at room temperature. Amylose-affinity chromatography was then used to separate cleaved mOLF variants. The mOLF variants were further polished by a final Superdex 75 followed by purity assessment with 12% resolving SDS-PAGE.

Thermal stability measurements
Each mOLF variant was tested for thermal stability by differential scanning fluorimetry as described previously (10). Briefly, purified mOLFs were buffer-exchanged into 10 mM HEPES, pH 7.5, 200 mM NaCl by concentrating and diluting three times using an Amicon 10,000 molecular-weight-cutoff filtration device. Reaction mixtures (30 l) containing 1ϫ SYPRO Orange and a final protein concentration of 3 M were prepared in triplicate at room temperature with or without 10 mM CaCl 2 . Samples were placed in a 96-well optical plate (Applied Biosystems) and sealed with optical film. Fluorescence measurements were obtained on an Applied Biosciences Step-One Plus realtime PCR instrument. Melts were conducted from 25 to 95°C with a 1°C/min increase. Data were analyzed using Origin software (OriginLab Corp.). The T m was calculated at the midpoint of unfolding using a Boltzmann sigmoid equation. Reported values are an average of two independent experiments.

CD
Near-UV CD measurements were acquired on a Jasco J-815 spectropolarimeter equipped with a Jasco PTC-4245/15 temperature control system. mOLF samples at a concentration range between 1.0 and 3.5 mg/ml were measured in gel filtra-tion buffer at 4°C. Scans were acquired from 250 to 320 nm at a rate of 50 nm/min and a data pitch of 1 nm using a 0.1-cm cuvette. Each measurement was an average of 20 scans. Data were blank-subtracted and converted to mean residue ellipticity ⌰ ϭ Mres ϫ ⌰obs/10 ϫ d ϫ c where Mres ϭ 112.9 is the mean residue mass calculated from the protein sequence, ⌰obs is the observed ellipticity (°) at wavelength , d is the pathlength (cm), and c is the protein concentration (g/ml). The reported spectra are an average of two independent measurements.

Intrinsic fluorescence measurements
Intrinsic fluorescence of purified mOLF at 1 M protein concentration in gel filtration was measured on a Shimadzu RF-5301PC spectrofluorophotometer at an excitation wavelength of 284 nm and an emission wavelength range of 300 -500 nm with 5-nm slit widths on excitation and emission monochromators. Spectra were collected nine times in 0.2-nm data intervals and averaged. Each variant was measured in triplicate. Reported values are an average of two independent experiments.

MD simulations
The WT mOLF structure (PDB code 4WXQ) (8) was used for MD simulations. The visualization and analysis program VMD (61) was used to solvate the system with ϳ14,500 TIP3P (62) water molecules in an 80 ϫ 75 ϫ 85-Å 3 box. The system was then neutralized with 0.15 M KCl, resulting in a final system size of ϳ48,000 atoms. Residue deletions and mutations were performed on the WT mOLF structure as described in the text.
MD simulations were carried out using NAMD 2.12 (63) and AMBER 16 (64) running on both CPUs and GPUs. The CHARMM36 force field was used for proteins (65). The temperature was maintained using Langevin dynamics; the pressure was kept at 1 atm using the Langevin piston method (66). The equations of motion were integrated using the RESPA multiple time-step algorithm with a time step of 2 fs for all bonded interactions, 2 fs for short-range nonbonded interactions, and 4 fs for long-range electrostatic interactions. Long-range electrostatic interactions were calculated using the particle-mesh Ewald method (67). Bonds involving hydrogen atoms were constrained to their equilibrium length. All mOLF systems were first allowed to equilibrate for 1 ns at 310 K with backbone atoms restrained before production simulations at 350 K. WT mOLF was simulated for 500 ns; all other mOLF variants were simulated for 100 ns.
Media from cells were collected and spun at 10,000 -16,000 ϫ g for 10 min. 200 l of media supernatant (Fig. 5A) or 7 l of media supernatant diluted into 100 l of nuclease-free water (Fig. 5B) was added into each well of a GE Healthcare dot blot apparatus and suctioned onto a nitrocellulose membrane. The membrane was then washed with PBS (filtered) twice and placed in Ponceau S to confirm the presence of protein followed by blocking with 5-7% milk for 1 h.
For intracellular fractionation to detect insoluble myocilin, cells were lysed either in 100 l of Triton X-100 lysis buffer (100 mM Tris-HCl, pH 7.4, 3 mM EGTA, 5 mM MgCl 2 , 0.5% Triton X-100) containing protease inhibitor mixture (Calbiochem) and 1ϫ phosphatase inhibitor II and III mixtures (Sigma) (Fig.  5A) or in 100 l of Mammalian Protein Extraction Reagent (M-PER) (Pierce) lysis buffer with 1ϫ cOmplete protease inhibitor mixture and 1ϫ phosphatase inhibitor II and III mixtures (Fig. 5B). Lysed cells were centrifuged at 10,000 -16,000 ϫ g for 10 min. Equal amounts of soluble supernatant from cell lysates were prepared with 2ϫ Laemmli sample buffer (Bio-Rad) containing 2-mercaptoethanol and denatured by boiling for 5 min at 100°C or 10 min at 95°C. For the insoluble faction presented in Fig. 5A, the pellet was washed with ice-cold PBS and then resuspended in 2ϫ Laemmli sample buffer with 9 M urea followed by sonication and denaturation by boiling. For Fig. 5B, the pellet was resuspended in equal volumes of Triton X-100 lysis buffer and 2ϫ Laemmli sample buffer with 2-mercaptoethanol and 8 M urea. Samples were then sonicated using a tip sonicator for 3 min with 10-s on/off pulses at 50% amplitude and then boiled in a 95°C water bath for 10 min. Prepared soluble and insoluble samples were then loaded onto a 10% Tris-glycine SDS-polyacrylamide gel. Gels were transferred onto polyvinylidene difluoride membranes (Millipore) and blocked for 1 h at room temperature with 5-7% milk prior to Western blotting.
A myocilin polyclonal antibody gifted from Dr. Dan Stamer (Duke University) was used to image blots presented in Fig. 5A